Forecasting Industrial Production in Brazil using many Predictors

Orientador(a): Marcelo Medeiros

In this article we compare the forecasting accuracy of standard and penalized regressions using many predictors on the brazilian industrial production index. We focus on the least absolute shrinkage and selection operator (LASSO) and its extensions. We also propose a combination between the LASSO and a deterministic variable search algorithm (LVSA). Models generated through a stochastic variable search algorithm (SVSA) were used as our benchmark specification. Our study produced three main findings. First, LASSO-based models underperform the benchmark in short- and long-term horizons. Second, the LVSA and the benchmark performances were virtually indistinguishable. Last, the best predictive variables are consistently chosen by all methods considered. As expected, these variables are closely related to Brazilian industrial activity. Examples include vehicle production and cardboard expedition.

MP 1